“…A general criterion for granule construction is to draw elements with indistinguishability, similarity, proximity or functionality together [28]. Traditional granulation methods adopt unsupervised clustering algorithms, such as Nearest Neighbour (NN) [26], K-Means [29], [30], hierarchical clustering [31], spatial partition trees [32], fuzzy C-means [20], [22], [33], [34], and prototype-based optimization [35], to construct granules. The unsupervised algorithm ensures the elements with certain similarity to be assembled into one granule, however the granules are non-interpretable in the view of semantic context.…”